Artificial Intelligence Nanodegree

Computer Vision Capstone

Project: Facial Keypoint Detection


Welcome to the final Computer Vision project in the Artificial Intelligence Nanodegree program!

In this project, you’ll combine your knowledge of computer vision techniques and deep learning to build and end-to-end facial keypoint recognition system! Facial keypoints include points around the eyes, nose, and mouth on any face and are used in many applications, from facial tracking to emotion recognition.

There are three main parts to this project:

Part 1 : Investigating OpenCV, pre-processing, and face detection

Part 2 : Training a Convolutional Neural Network (CNN) to detect facial keypoints

Part 3 : Putting parts 1 and 2 together to identify facial keypoints on any image!


*Here's what you need to know to complete the project:

  1. In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested.

    a. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

  1. In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation.

    a. Each section where you will answer a question is preceded by a 'Question X' header.

    b. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional suggestions for enhancing the project beyond the minimum requirements. If you decide to pursue the "(Optional)" sections, you should include the code in this IPython notebook.

Your project submission will be evaluated based on your answers to each of the questions and the code implementations you provide.

Steps to Complete the Project

Each part of the notebook is further broken down into separate steps. Feel free to use the links below to navigate the notebook.

In this project you will get to explore a few of the many computer vision algorithms built into the OpenCV library. This expansive computer vision library is now almost 20 years old and still growing!

The project itself is broken down into three large parts, then even further into separate steps. Make sure to read through each step, and complete any sections that begin with '(IMPLEMENTATION)' in the header; these implementation sections may contain multiple TODOs that will be marked in code. For convenience, we provide links to each of these steps below.

Part 1 : Investigating OpenCV, pre-processing, and face detection

  • Step 0: Detect Faces Using a Haar Cascade Classifier
  • Step 1: Add Eye Detection
  • Step 2: De-noise an Image for Better Face Detection
  • Step 3: Blur an Image and Perform Edge Detection
  • Step 4: Automatically Hide the Identity of an Individual

Part 2 : Training a Convolutional Neural Network (CNN) to detect facial keypoints

  • Step 5: Create a CNN to Recognize Facial Keypoints
  • Step 6: Compile and Train the Model
  • Step 7: Visualize the Loss and Answer Questions

Part 3 : Putting parts 1 and 2 together to identify facial keypoints on any image!

  • Step 8: Build a Robust Facial Keypoints Detector (Complete the CV Pipeline)

Step 0: Detect Faces Using a Haar Cascade Classifier

Have you ever wondered how Facebook automatically tags images with your friends' faces? Or how high-end cameras automatically find and focus on a certain person's face? Applications like these depend heavily on the machine learning task known as face detection - which is the task of automatically finding faces in images containing people.

At its root face detection is a classification problem - that is a problem of distinguishing between distinct classes of things. With face detection these distinct classes are 1) images of human faces and 2) everything else.

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the detector_architectures directory.

Import Resources

In the next python cell, we load in the required libraries for this section of the project.

In [1]:
# Import required libraries for this section

%matplotlib inline

import numpy as np
import matplotlib.pyplot as plt
import math
import cv2                     # OpenCV library for computer vision
from PIL import Image
import time 

Next, we load in and display a test image for performing face detection.

Note: by default OpenCV assumes the ordering of our image's color channels are Blue, then Green, then Red. This is slightly out of order with most image types we'll use in these experiments, whose color channels are ordered Red, then Green, then Blue. In order to switch the Blue and Red channels of our test image around we will use OpenCV's cvtColor function, which you can read more about by checking out some of its documentation located here. This is a general utility function that can do other transformations too like converting a color image to grayscale, and transforming a standard color image to HSV color space.

In [3]:
# Load in color image for face detection
image = cv2.imread('images/test_image_1.jpg')

# Convert the image to RGB colorspace
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Plot our image using subplots to specify a size and title
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Original Image')
ax1.imshow(image)
Out[3]:
<matplotlib.image.AxesImage at 0x17c7c019860>

There are a lot of people - and faces - in this picture. 13 faces to be exact! In the next code cell, we demonstrate how to use a Haar Cascade classifier to detect all the faces in this test image.

This face detector uses information about patterns of intensity in an image to reliably detect faces under varying light conditions. So, to use this face detector, we'll first convert the image from color to grayscale.

Then, we load in the fully trained architecture of the face detector -- found in the file haarcascade_frontalface_default.xml - and use it on our image to find faces!

To learn more about the parameters of the detector see this post.

In [4]:
# Convert the RGB  image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)

# Extract the pre-trained face detector from an xml file
face_cascade = cv2.CascadeClassifier('detector_architectures/haarcascade_frontalface_default.xml')

# Detect the faces in image
faces = face_cascade.detectMultiScale(gray, 4, 6)

# Print the number of faces detected in the image
print('Number of faces detected:', len(faces))

# Make a copy of the orginal image to draw face detections on
image_with_detections = np.copy(image)

# Get the bounding box for each detected face
for (x,y,w,h) in faces:
    # Add a red bounding box to the detections image
    cv2.rectangle(image_with_detections, (x,y), (x+w,y+h), (255,0,0), 3)
    

# Display the image with the detections
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Image with Face Detections')
ax1.imshow(image_with_detections)
Number of faces detected: 13
Out[4]:
<matplotlib.image.AxesImage at 0x17c7ca97898>

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.


Step 1: Add Eye Detections

There are other pre-trained detectors available that use a Haar Cascade Classifier - including full human body detectors, license plate detectors, and more. A full list of the pre-trained architectures can be found here.

To test your eye detector, we'll first read in a new test image with just a single face.

In [5]:
# Load in color image for face detection
image = cv2.imread('images/james.jpg')

# Convert the image to RGB colorspace
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Plot the RGB image
fig = plt.figure(figsize = (6,6))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Original Image')
ax1.imshow(image)
Out[5]:
<matplotlib.image.AxesImage at 0x17c023286a0>

Notice that even though the image is a black and white image, we have read it in as a color image and so it will still need to be converted to grayscale in order to perform the most accurate face detection.

So, the next steps will be to convert this image to grayscale, then load OpenCV's face detector and run it with parameters that detect this face accurately.

In [6]:
# Convert the RGB  image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)

# Extract the pre-trained face detector from an xml file
face_cascade = cv2.CascadeClassifier('detector_architectures/haarcascade_frontalface_default.xml')

# Detect the faces in image
faces = face_cascade.detectMultiScale(gray, 1.25, 6)

# Print the number of faces detected in the image
print('Number of faces detected:', len(faces))

# Make a copy of the orginal image to draw face detections on
image_with_detections = np.copy(image)

# Get the bounding box for each detected face
for (x,y,w,h) in faces:
    # Add a red bounding box to the detections image
    cv2.rectangle(image_with_detections, (x,y), (x+w,y+h), (255,0,0), 3)
    

# Display the image with the detections
fig = plt.figure(figsize = (6,6))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Image with Face Detection')
ax1.imshow(image_with_detections)
Number of faces detected: 1
Out[6]:
<matplotlib.image.AxesImage at 0x17c0032cb70>

(IMPLEMENTATION) Add an eye detector to the current face detection setup.

A Haar-cascade eye detector can be included in the same way that the face detector was and, in this first task, it will be your job to do just this.

To set up an eye detector, use the stored parameters of the eye cascade detector, called haarcascade_eye.xml, located in the detector_architectures subdirectory. In the next code cell, create your eye detector and store its detections.

A few notes before you get started:

First, make sure to give your loaded eye detector the variable name

eye_cascade

and give the list of eye regions you detect the variable name

eyes

Second, since we've already run the face detector over this image, you should only search for eyes within the rectangular face regions detected in faces. This will minimize false detections.

Lastly, once you've run your eye detector over the facial detection region, you should display the RGB image with both the face detection boxes (in red) and your eye detections (in green) to verify that everything works as expected.

In [7]:
# Make a copy of the original image to plot rectangle detections
image_with_detections = np.copy(image)

# Loop over the detections and draw their corresponding face detection boxes
for (x,y,w,h) in faces:
    cv2.rectangle(image_with_detections, (x,y), (x+w,y+h),(255,0,0), 3)  
    
# Do not change the code above this comment!
    
## TODO: Add eye detection, using haarcascade_eye.xml, to the current face detector algorithm
eye_cascade = cv2.CascadeClassifier('detector_architectures/haarcascade_eye.xml')

for (x,y,w,h) in faces:
    roi_gray = gray[y:y+h, x:x+w]
    roi_color = image_with_detections[y:y+h, x:x+w]
    eyes = eye_cascade.detectMultiScale(roi_gray)
    
## TODO: Loop over the eye detections and draw their corresponding boxes in green on image_with_detections
    for (ex,ey,ew,eh) in eyes:
        cv2.rectangle(roi_color,(ex,ey),(ex+ew,ey+eh),(0,255,0),2)

# Plot the image with both faces and eyes detected
fig = plt.figure(figsize = (6,6))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Image with Face and Eye Detection')
ax1.imshow(image_with_detections)
Out[7]:
<matplotlib.image.AxesImage at 0x17c004ddf60>

(Optional) Add face and eye detection to your laptop camera

It's time to kick it up a notch, and add face and eye detection to your laptop's camera! Afterwards, you'll be able to show off your creation like in the gif shown below - made with a completed version of the code!

Notice that not all of the detections here are perfect - and your result need not be perfect either. You should spend a small amount of time tuning the parameters of your detectors to get reasonable results, but don't hold out for perfection. If we wanted perfection we'd need to spend a ton of time tuning the parameters of each detector, cleaning up the input image frames, etc. You can think of this as more of a rapid prototype.

The next cell contains code for a wrapper function called laptop_camera_face_eye_detector that, when called, will activate your laptop's camera. You will place the relevant face and eye detection code in this wrapper function to implement face/eye detection and mark those detections on each image frame that your camera captures.

Before adding anything to the function, you can run it to get an idea of how it works - a small window should pop up showing you the live feed from your camera; you can press any key to close this window.

Note: Mac users may find that activating this function kills the kernel of their notebook every once in a while. If this happens to you, just restart your notebook's kernel, activate cell(s) containing any crucial import statements, and you'll be good to go!

In [7]:
### Add face and eye detection to this laptop camera function 
# Make sure to draw out all faces/eyes found in each frame on the shown video feed

import cv2
import time 

# wrapper function for face/eye detection with your laptop camera
def laptop_camera_go():
    # Create instance of video capturer
    cv2.namedWindow("face detection activated")
    vc = cv2.VideoCapture(0)

    # Try to get the first frame
    if vc.isOpened(): 
        rval, frame = vc.read()
    else:
        rval = False
    
    # Keep the video stream open
    while rval:
        # Plot the image from camera with all the face and eye detections marked
        cv2.imshow("face detection activated", frame)
        
        # Exit functionality - press any key to exit laptop video
        key = cv2.waitKey(20)
        if key > 0: # Exit by pressing any key
            # Destroy windows 
            cv2.destroyAllWindows()
            
            # Make sure window closes on OSx
            for i in range (1,5):
                cv2.waitKey(1)
            return
        
        # Read next frame
        time.sleep(0.05)             # control framerate for computation - default 20 frames per sec
        rval, frame = vc.read()    
In [ ]:
# Call the laptop camera face/eye detector function above
laptop_camera_go()

Step 2: De-noise an Image for Better Face Detection

Image quality is an important aspect of any computer vision task. Typically, when creating a set of images to train a deep learning network, significant care is taken to ensure that training images are free of visual noise or artifacts that hinder object detection. While computer vision algorithms - like a face detector - are typically trained on 'nice' data such as this, new test data doesn't always look so nice!

When applying a trained computer vision algorithm to a new piece of test data one often cleans it up first before feeding it in. This sort of cleaning - referred to as pre-processing - can include a number of cleaning phases like blurring, de-noising, color transformations, etc., and many of these tasks can be accomplished using OpenCV.

In this short subsection we explore OpenCV's noise-removal functionality to see how we can clean up a noisy image, which we then feed into our trained face detector.

Create a noisy image to work with

In the next cell, we create an artificial noisy version of the previous multi-face image. This is a little exaggerated - we don't typically get images that are this noisy - but image noise, or 'grainy-ness' in a digitial image - is a fairly common phenomenon.

In [11]:
# Load in the multi-face test image again
image = cv2.imread('images/test_image_1.jpg')

# Convert the image copy to RGB colorspace
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Make an array copy of this image
image_with_noise = np.asarray(image)

# Create noise - here we add noise sampled randomly from a Gaussian distribution: a common model for noise
noise_level = 40
noise = np.random.randn(image.shape[0],image.shape[1],image.shape[2])*noise_level

# Add this noise to the array image copy
image_with_noise = image_with_noise + noise

# Convert back to uint8 format
image_with_noise = np.asarray([np.uint8(np.clip(i,0,255)) for i in image_with_noise])

# Plot our noisy image!
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Noisy Image')
ax1.imshow(image_with_noise)
Out[11]:
<matplotlib.image.AxesImage at 0x21088e26ac8>

In the context of face detection, the problem with an image like this is that - due to noise - we may miss some faces or get false detections.

In the next cell we apply the same trained OpenCV detector with the same settings as before, to see what sort of detections we get.

In [12]:
# Convert the RGB  image to grayscale
gray_noise = cv2.cvtColor(image_with_noise, cv2.COLOR_RGB2GRAY)

# Extract the pre-trained face detector from an xml file
face_cascade = cv2.CascadeClassifier('detector_architectures/haarcascade_frontalface_default.xml')

# Detect the faces in image
faces = face_cascade.detectMultiScale(gray_noise, 4, 6)

# Print the number of faces detected in the image
print('Number of faces detected:', len(faces))

# Make a copy of the orginal image to draw face detections on
image_with_detections = np.copy(image_with_noise)

# Get the bounding box for each detected face
for (x,y,w,h) in faces:
    # Add a red bounding box to the detections image
    cv2.rectangle(image_with_detections, (x,y), (x+w,y+h), (255,0,0), 3)
    

# Display the image with the detections
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Noisy Image with Face Detections')
ax1.imshow(image_with_detections)
Number of faces detected: 12
Out[12]:
<matplotlib.image.AxesImage at 0x21088b4d320>

With this added noise we now miss one of the faces!

(IMPLEMENTATION) De-noise this image for better face detection

Time to get your hands dirty: using OpenCV's built in color image de-noising functionality called fastNlMeansDenoisingColored - de-noise this image enough so that all the faces in the image are properly detected. Once you have cleaned the image in the next cell, use the cell that follows to run our trained face detector over the cleaned image to check out its detections.

You can find its official documentation here and a useful example here.

Note: you can keep all parameters except photo_render fixed as shown in the second link above. Play around with the value of this parameter - see how it affects the resulting cleaned image.

In [24]:
## TODO: Use OpenCV's built in color image de-noising function to clean up our noisy image!
denoised_image = cv2.fastNlMeansDenoisingColored(image_with_noise,None,17,17,7,21)

# your final de-noised image (should be RGB)

# Display the de-noised image
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('De-Noised Image')
ax1.imshow(denoised_image)
Out[24]:
<matplotlib.image.AxesImage at 0x21088ff2be0>
In [25]:
## TODO: Run the face detector on the de-noised image to improve your detections and display the result

# Convert the RGB  image to grayscale
gray_denoised = cv2.cvtColor(denoised_image, cv2.COLOR_RGB2GRAY)

# Extract the pre-trained face detector from an xml file
face_cascade = cv2.CascadeClassifier('detector_architectures/haarcascade_frontalface_default.xml')

# Detect the faces in image
faces = face_cascade.detectMultiScale(gray_denoised, 4, 6)

# Print the number of faces detected in the image
print('Number of faces detected:', len(faces))

# Make a copy of the orginal image to draw face detections on
denoised_image_with_detections = np.copy(denoised_image)

# Get the bounding box for each detected face
for (x,y,w,h) in faces:
    # Add a red bounding box to the detections image
    cv2.rectangle(denoised_image_with_detections, (x,y), (x+w,y+h), (255,0,0), 3)
    

# Display the image with the detections
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Noisy Image with Face Detections')
ax1.imshow(denoised_image_with_detections)
Number of faces detected: 13
Out[25]:
<matplotlib.image.AxesImage at 0x210890442b0>

Step 3: Blur an Image and Perform Edge Detection

Now that we have developed a simple pipeline for detecting faces using OpenCV - let's start playing around with a few fun things we can do with all those detected faces!

Importance of Blur in Edge Detection

Edge detection is a concept that pops up almost everywhere in computer vision applications, as edge-based features (as well as features built on top of edges) are often some of the best features for e.g., object detection and recognition problems.

Edge detection is a dimension reduction technique - by keeping only the edges of an image we get to throw away a lot of non-discriminating information. And typically the most useful kind of edge-detection is one that preserves only the important, global structures (ignoring local structures that aren't very discriminative). So removing local structures / retaining global structures is a crucial pre-processing step to performing edge detection in an image, and blurring can do just that.

Below is an animated gif showing the result of an edge-detected cat taken from Wikipedia, where the image is gradually blurred more and more prior to edge detection. When the animation begins you can't quite make out what it's a picture of, but as the animation evolves and local structures are removed via blurring the cat becomes visible in the edge-detected image.

Edge detection is a convolution performed on the image itself, and you can read about Canny edge detection on this OpenCV documentation page.

Canny edge detection

In the cell below we load in a test image, then apply Canny edge detection on it. The original image is shown on the left panel of the figure, while the edge-detected version of the image is shown on the right. Notice how the result looks very busy - there are too many little details preserved in the image before it is sent to the edge detector. When applied in computer vision applications, edge detection should preserve global structure; doing away with local structures that don't help describe what objects are in the image.

In [17]:
# Load in the image
image = cv2.imread('images/fawzia.jpg')

# Convert to RGB colorspace
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Convert to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)  

# Perform Canny edge detection
edges = cv2.Canny(gray,100,200)

# Dilate the image to amplify edges
edges = cv2.dilate(edges, None)

# Plot the RGB and edge-detected image
fig = plt.figure(figsize = (15,15))
ax1 = fig.add_subplot(121)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Original Image')
ax1.imshow(image)

ax2 = fig.add_subplot(122)
ax2.set_xticks([])
ax2.set_yticks([])

ax2.set_title('Canny Edges')
ax2.imshow(edges, cmap='gray')
Out[17]:
<matplotlib.image.AxesImage at 0x188488c0710>

Without first blurring the image, and removing small, local structures, a lot of irrelevant edge content gets picked up and amplified by the detector (as shown in the right panel above).

(IMPLEMENTATION) Blur the image then perform edge detection

In the next cell, you will repeat this experiment - blurring the image first to remove these local structures, so that only the important boudnary details remain in the edge-detected image.

Blur the image by using OpenCV's filter2d functionality - which is discussed in this documentation page - and use an averaging kernel of width equal to 4.

In [27]:
### TODO: Blur the test imageusing OpenCV's filter2d functionality, 
# Use an averaging kernel, and a kernel width equal to 4


# Load in the image
image = cv2.imread('images/fawzia.jpg')

# Convert to RGB colorspace
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Convert to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)

# Blur Image
kernel = np.ones((4,4),np.float32)/16
blurred_image = cv2.filter2D(gray,-1,kernel)

## TODO: Then perform Canny edge detection and display the output

# Perform Canny edge detection
edges = cv2.Canny(blurred_image,100,200)

# Dilate the image to amplify edges
edges = cv2.dilate(edges, None)

# Plot the RGB and edge-detected image
fig = plt.figure(figsize = (15,15))
ax1 = fig.add_subplot(121)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Original Image')
ax1.imshow(image)

ax2 = fig.add_subplot(122)
ax2.set_xticks([])
ax2.set_yticks([])

ax2.set_title('Canny Edges')
ax2.imshow(edges, cmap='gray')
Out[27]:
<matplotlib.image.AxesImage at 0x210892974e0>

Step 4: Automatically Hide the Identity of an Individual

If you film something like a documentary or reality TV, you must get permission from every individual shown on film before you can show their face, otherwise you need to blur it out - by blurring the face a lot (so much so that even the global structures are obscured)! This is also true for projects like Google's StreetView maps - an enormous collection of mapping images taken from a fleet of Google vehicles. Because it would be impossible for Google to get the permission of every single person accidentally captured in one of these images they blur out everyone's faces, the detected images must automatically blur the identity of detected people. Here's a few examples of folks caught in the camera of a Google street view vehicle.

Read in an image to perform identity detection

Let's try this out for ourselves. Use the face detection pipeline built above and what you know about using the filter2D to blur and image, and use these in tandem to hide the identity of the person in the following image - loaded in and printed in the next cell.

In [28]:
# Load in the image
image = cv2.imread('images/gus.jpg')

# Convert the image to RGB colorspace
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Display the image
fig = plt.figure(figsize = (6,6))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])
ax1.set_title('Original Image')
ax1.imshow(image)
Out[28]:
<matplotlib.image.AxesImage at 0x210892d6710>

(IMPLEMENTATION) Use blurring to hide the identity of an individual in an image

The idea here is to 1) automatically detect the face in this image, and then 2) blur it out! Make sure to adjust the parameters of the averaging blur filter to completely obscure this person's identity.

In [33]:
## TODO: Implement face detection

# Extract the pre-trained face detector from an xml file
face_cascade = cv2.CascadeClassifier('detector_architectures/haarcascade_frontalface_default.xml')

# Detect the faces in image
faces = face_cascade.detectMultiScale(image, 4, 6)

# Make a copy of the orginal image to draw face detections on
image_with_blurred_faces = np.copy(image)

# Get the bounding box for each detected face
for (x,y,w,h) in faces:
    # Add a red bounding box to the detections image
     image_with_blurred_faces[y:y+h, x:x+w] = cv2.blur(image_with_blurred_faces[y:y+h, x:x+w],(100,100))

## TODO: Blur the bounding box around each detected face using an averaging filter and display the result

# Display the image with the blurred faces
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])

ax1.set_title('Image with blurred faces')
ax1.imshow(image_with_blurred_faces)
Out[33]:
<matplotlib.image.AxesImage at 0x21088999f98>

(Optional) Build identity protection into your laptop camera

In this optional task you can add identity protection to your laptop camera, using the previously completed code where you added face detection to your laptop camera - and the task above. You should be able to get reasonable results with little parameter tuning - like the one shown in the gif below.

As with the previous video task, to make this perfect would require significant effort - so don't strive for perfection here, strive for reasonable quality.

The next cell contains code a wrapper function called laptop_camera_identity_hider that - when called - will activate your laptop's camera. You need to place the relevant face detection and blurring code developed above in this function in order to blur faces entering your laptop camera's field of view.

Before adding anything to the function you can call it to get a hang of how it works - a small window will pop up showing you the live feed from your camera, you can press any key to close this window.

Note: Mac users may find that activating this function kills the kernel of their notebook every once in a while. If this happens to you, just restart your notebook's kernel, activate cell(s) containing any crucial import statements, and you'll be good to go!

In [ ]:
### Insert face detection and blurring code into the wrapper below to create an identity protector on your laptop!
import cv2
import time 

def laptop_camera_go():
    # Create instance of video capturer
    cv2.namedWindow("face detection activated")
    vc = cv2.VideoCapture(0)

    # Try to get the first frame
    if vc.isOpened(): 
        rval, frame = vc.read()
    else:
        rval = False
    
    # Keep video stream open
    while rval:
        # Plot image from camera with detections marked
        cv2.imshow("face detection activated", frame)
        
        # Exit functionality - press any key to exit laptop video
        key = cv2.waitKey(20)
        if key > 0: # Exit by pressing any key
            # Destroy windows
            cv2.destroyAllWindows()
            
            for i in range (1,5):
                cv2.waitKey(1)
            return
        
        # Read next frame
        time.sleep(0.05)             # control framerate for computation - default 20 frames per sec
        rval, frame = vc.read()    
        
In [ ]:
# Run laptop identity hider
laptop_camera_go()

Step 5: Create a CNN to Recognize Facial Keypoints

OpenCV is often used in practice with other machine learning and deep learning libraries to produce interesting results. In this stage of the project you will create your own end-to-end pipeline - employing convolutional networks in keras along with OpenCV - to apply a "selfie" filter to streaming video and images.

You will start by creating and then training a convolutional network that can detect facial keypoints in a small dataset of cropped images of human faces. We then guide you towards OpenCV to expanding your detection algorithm to more general images. What are facial keypoints? Let's take a look at some examples.

Facial keypoints (also called facial landmarks) are the small blue-green dots shown on each of the faces in the image above - there are 15 keypoints marked in each image. They mark important areas of the face - the eyes, corners of the mouth, the nose, etc. Facial keypoints can be used in a variety of machine learning applications from face and emotion recognition to commercial applications like the image filters popularized by Snapchat.

Below we illustrate a filter that, using the results of this section, automatically places sunglasses on people in images (using the facial keypoints to place the glasses correctly on each face). Here, the facial keypoints have been colored lime green for visualization purposes.

Make a facial keypoint detector

But first things first: how can we make a facial keypoint detector? Well, at a high level, notice that facial keypoint detection is a regression problem. A single face corresponds to a set of 15 facial keypoints (a set of 15 corresponding $(x, y)$ coordinates, i.e., an output point). Because our input data are images, we can employ a convolutional neural network to recognize patterns in our images and learn how to identify these keypoint given sets of labeled data.

In order to train a regressor, we need a training set - a set of facial image / facial keypoint pairs to train on. For this we will be using this dataset from Kaggle. We've already downloaded this data and placed it in the data directory. Make sure that you have both the training and test data files. The training dataset contains several thousand $96 \times 96$ grayscale images of cropped human faces, along with each face's 15 corresponding facial keypoints (also called landmarks) that have been placed by hand, and recorded in $(x, y)$ coordinates. This wonderful resource also has a substantial testing set, which we will use in tinkering with our convolutional network.

To load in this data, run the Python cell below - notice we will load in both the training and testing sets.

The load_data function is in the included utils.py file.

In [6]:
from utils import *

# Load training set
X_train, y_train = load_data()
print("X_train.shape == {}".format(X_train.shape))
print("y_train.shape == {}; y_train.min == {:.3f}; y_train.max == {:.3f}".format(
    y_train.shape, y_train.min(), y_train.max()))

# Load testing set
X_test, _ = load_data(test=True)
print("X_test.shape == {}".format(X_test.shape))
X_train.shape == (2140, 96, 96, 1)
y_train.shape == (2140, 30); y_train.min == -0.920; y_train.max == 0.996
X_test.shape == (1783, 96, 96, 1)

The load_data function in utils.py originates from this excellent blog post, which you are strongly encouraged to read. Please take the time now to review this function. Note how the output values - that is, the coordinates of each set of facial landmarks - have been normalized to take on values in the range $[-1, 1]$, while the pixel values of each input point (a facial image) have been normalized to the range $[0,1]$.

Note: the original Kaggle dataset contains some images with several missing keypoints. For simplicity, the load_data function removes those images with missing labels from the dataset. As an optional extension, you are welcome to amend the load_data function to include the incomplete data points.

Visualize the Training Data

Execute the code cell below to visualize a subset of the training data.

In [55]:
import matplotlib.pyplot as plt
%matplotlib inline

fig = plt.figure(figsize=(20,20))
fig.subplots_adjust(left=0, right=1, bottom=0, top=1, hspace=0.05, wspace=0.05)
for i in range(9):
    ax = fig.add_subplot(3, 3, i + 1, xticks=[], yticks=[])
    plot_data(X_train[i], y_train[i], ax)

For each training image, there are two landmarks per eyebrow (four total), three per eye (six total), four for the mouth, and one for the tip of the nose.

Review the plot_data function in utils.py to understand how the 30-dimensional training labels in y_train are mapped to facial locations, as this function will prove useful for your pipeline.

(IMPLEMENTATION) Specify the CNN Architecture

In this section, you will specify a neural network for predicting the locations of facial keypoints. Use the code cell below to specify the architecture of your neural network. We have imported some layers that you may find useful for this task, but if you need to use more Keras layers, feel free to import them in the cell.

Your network should accept a $96 \times 96$ grayscale image as input, and it should output a vector with 30 entries, corresponding to the predicted (horizontal and vertical) locations of 15 facial keypoints. If you are not sure where to start, you can find some useful starting architectures in this blog, but you are not permitted to copy any of the architectures that you find online.

In [179]:
# Import deep learning resources from Keras
from keras.models import Sequential
from keras.layers import Convolution2D, MaxPooling2D, Dropout
from keras.layers import Flatten, Dense


## TODO: Specify a CNN architecture
# Your model should accept 96x96 pixel graysale images in
# It should have a fully-connected output layer with 30 values (2 for each facial keypoint)

model_architecture = Sequential()

model_architecture.add(Convolution2D(32, (3, 3), padding='same', input_shape=(X_train.shape[1:]), activation='relu'))
model_architecture.add(Convolution2D(32, (3, 3), padding='same', activation='relu'))
model_architecture.add(MaxPooling2D(pool_size=(2, 2)))
model_architecture.add(Dropout(0.25))

model_architecture.add(Convolution2D(64, (3, 3), padding='same', activation='relu'))
model_architecture.add(Convolution2D(64, (3, 3), padding='same', activation='relu'))
model_architecture.add(MaxPooling2D(pool_size=(2, 2)))
model_architecture.add(Dropout(0.25))

model_architecture.add(Convolution2D(128, (3, 3), padding='same', activation='relu'))
model_architecture.add(MaxPooling2D(pool_size=(2, 2)))
model_architecture.add(Convolution2D(128, (3, 3), padding='same', activation='relu'))
model_architecture.add(MaxPooling2D(pool_size=(2, 2)))
model_architecture.add(Dropout(0.25))

model_architecture.add(Flatten())
model_architecture.add(Dense(1024, activation='relu'))
model_architecture.add(Dropout(0.5))
model_architecture.add(Dense(y_train.shape[-1]))

# Summarize the model
model_architecture.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_299 (Conv2D)          (None, 96, 96, 32)        320       
_________________________________________________________________
conv2d_300 (Conv2D)          (None, 96, 96, 32)        9248      
_________________________________________________________________
max_pooling2d_189 (MaxPoolin (None, 48, 48, 32)        0         
_________________________________________________________________
dropout_232 (Dropout)        (None, 48, 48, 32)        0         
_________________________________________________________________
conv2d_301 (Conv2D)          (None, 48, 48, 64)        18496     
_________________________________________________________________
conv2d_302 (Conv2D)          (None, 48, 48, 64)        36928     
_________________________________________________________________
max_pooling2d_190 (MaxPoolin (None, 24, 24, 64)        0         
_________________________________________________________________
dropout_233 (Dropout)        (None, 24, 24, 64)        0         
_________________________________________________________________
conv2d_303 (Conv2D)          (None, 24, 24, 128)       73856     
_________________________________________________________________
max_pooling2d_191 (MaxPoolin (None, 12, 12, 128)       0         
_________________________________________________________________
conv2d_304 (Conv2D)          (None, 12, 12, 128)       147584    
_________________________________________________________________
max_pooling2d_192 (MaxPoolin (None, 6, 6, 128)         0         
_________________________________________________________________
dropout_234 (Dropout)        (None, 6, 6, 128)         0         
_________________________________________________________________
flatten_65 (Flatten)         (None, 4608)              0         
_________________________________________________________________
dense_139 (Dense)            (None, 1024)              4719616   
_________________________________________________________________
dropout_235 (Dropout)        (None, 1024)              0         
_________________________________________________________________
dense_140 (Dense)            (None, 30)                30750     
=================================================================
Total params: 5,036,798
Trainable params: 5,036,798
Non-trainable params: 0
_________________________________________________________________

Step 6: Compile and Train the Model

After specifying your architecture, you'll need to compile and train the model to detect facial keypoints'

(IMPLEMENTATION) Compile and Train the Model

Use the compile method to configure the learning process. Experiment with your choice of optimizer; you may have some ideas about which will work best (SGD vs. RMSprop, etc), but take the time to empirically verify your theories.

Use the fit method to train the model. Break off a validation set by setting validation_split=0.2. Save the returned History object in the history variable.

Experiment with your model to minimize the validation loss (measured as mean squared error). A very good model will achieve about 0.0015 loss (though it's possible to do even better). When you have finished training, save your model as an HDF5 file with file path my_model.h5.

In [180]:
from keras.optimizers import SGD, RMSprop, Adagrad, Adadelta, Adam, Adamax, Nadam
from keras.callbacks import ModelCheckpoint, EarlyStopping
import time

model = model_architecture
loss_function = 'mean_squared_error'
optimizer = 'adamax'
detail = 'updated_best_model_' + loss_function +'_'+ optimizer 

## TODO: Compile the model
model.compile(loss=loss_function, optimizer=optimizer)

## TODO: Save the best model as best_model.h5 for checkpoints
checkpointer = ModelCheckpoint(filepath='saved_models/{}_{}.h5'.format(detail, time.time()), verbose=1, save_best_only=True)
early_stopping = EarlyStopping(monitor='val_loss', patience=30)

## TODO: Train the model
hist = model.fit(X_train, y_train, batch_size=64, epochs=1000, validation_split=0.2, callbacks=[checkpointer, early_stopping], verbose=1)
Train on 1712 samples, validate on 428 samples
Epoch 1/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0400Epoch 00000: val_loss improved from inf to 0.00788, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 7s - loss: 0.0392 - val_loss: 0.0079
Epoch 2/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0093Epoch 00001: val_loss improved from 0.00788 to 0.00671, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0093 - val_loss: 0.0067
Epoch 3/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0080Epoch 00002: val_loss improved from 0.00671 to 0.00545, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0080 - val_loss: 0.0054
Epoch 4/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0076Epoch 00003: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0076 - val_loss: 0.0058
Epoch 5/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0072Epoch 00004: val_loss improved from 0.00545 to 0.00537, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0072 - val_loss: 0.0054
Epoch 6/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0070Epoch 00005: val_loss improved from 0.00537 to 0.00500, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0070 - val_loss: 0.0050
Epoch 7/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0067Epoch 00006: val_loss improved from 0.00500 to 0.00470, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0067 - val_loss: 0.0047
Epoch 8/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0066Epoch 00007: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0066 - val_loss: 0.0050
Epoch 9/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0063Epoch 00008: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0063 - val_loss: 0.0049
Epoch 10/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0062Epoch 00009: val_loss improved from 0.00470 to 0.00445, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0062 - val_loss: 0.0045
Epoch 11/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0060Epoch 00010: val_loss improved from 0.00445 to 0.00431, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0060 - val_loss: 0.0043
Epoch 12/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0058Epoch 00011: val_loss improved from 0.00431 to 0.00412, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0058 - val_loss: 0.0041
Epoch 13/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0058Epoch 00012: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0058 - val_loss: 0.0045
Epoch 14/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0057Epoch 00013: val_loss improved from 0.00412 to 0.00412, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0057 - val_loss: 0.0041
Epoch 15/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0056Epoch 00014: val_loss improved from 0.00412 to 0.00397, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0056 - val_loss: 0.0040
Epoch 16/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0054Epoch 00015: val_loss improved from 0.00397 to 0.00387, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0054 - val_loss: 0.0039
Epoch 17/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0052Epoch 00016: val_loss improved from 0.00387 to 0.00362, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0052 - val_loss: 0.0036
Epoch 18/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0050Epoch 00017: val_loss improved from 0.00362 to 0.00352, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0050 - val_loss: 0.0035
Epoch 19/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0049Epoch 00018: val_loss improved from 0.00352 to 0.00340, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0049 - val_loss: 0.0034
Epoch 20/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0049Epoch 00019: val_loss improved from 0.00340 to 0.00326, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0049 - val_loss: 0.0033
Epoch 21/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0047Epoch 00020: val_loss improved from 0.00326 to 0.00321, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0046 - val_loss: 0.0032
Epoch 22/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0042Epoch 00021: val_loss improved from 0.00321 to 0.00277, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0042 - val_loss: 0.0028
Epoch 23/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0042Epoch 00022: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0042 - val_loss: 0.0032
Epoch 24/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0039Epoch 00023: val_loss improved from 0.00277 to 0.00236, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0039 - val_loss: 0.0024
Epoch 25/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0039Epoch 00024: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0039 - val_loss: 0.0029
Epoch 26/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0038Epoch 00025: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0038 - val_loss: 0.0027
Epoch 27/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0036Epoch 00026: val_loss improved from 0.00236 to 0.00212, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0036 - val_loss: 0.0021
Epoch 28/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0034Epoch 00027: val_loss improved from 0.00212 to 0.00195, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0034 - val_loss: 0.0019
Epoch 29/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0034Epoch 00028: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0034 - val_loss: 0.0021
Epoch 30/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0032Epoch 00029: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0033 - val_loss: 0.0021
Epoch 31/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0033Epoch 00030: val_loss improved from 0.00195 to 0.00193, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0033 - val_loss: 0.0019
Epoch 32/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0032Epoch 00031: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0032 - val_loss: 0.0019
Epoch 33/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0031Epoch 00032: val_loss improved from 0.00193 to 0.00186, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0031 - val_loss: 0.0019
Epoch 34/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0030Epoch 00033: val_loss improved from 0.00186 to 0.00166, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0030 - val_loss: 0.0017
Epoch 35/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0030Epoch 00034: val_loss improved from 0.00166 to 0.00164, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0030 - val_loss: 0.0016
Epoch 36/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0029Epoch 00035: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0030 - val_loss: 0.0018
Epoch 37/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0029Epoch 00036: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0029 - val_loss: 0.0017
Epoch 38/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0029Epoch 00037: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0028 - val_loss: 0.0019
Epoch 39/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0027Epoch 00038: val_loss improved from 0.00164 to 0.00155, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0027 - val_loss: 0.0016
Epoch 40/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0027Epoch 00039: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0028 - val_loss: 0.0018
Epoch 41/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0028Epoch 00040: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0028 - val_loss: 0.0018
Epoch 42/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0027Epoch 00041: val_loss improved from 0.00155 to 0.00155, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0027 - val_loss: 0.0015
Epoch 43/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0026Epoch 00042: val_loss improved from 0.00155 to 0.00144, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0026 - val_loss: 0.0014
Epoch 44/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0025Epoch 00043: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0025 - val_loss: 0.0015
Epoch 45/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0024Epoch 00044: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0024 - val_loss: 0.0015
Epoch 46/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0025Epoch 00045: val_loss improved from 0.00144 to 0.00138, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0025 - val_loss: 0.0014
Epoch 47/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0024Epoch 00046: val_loss improved from 0.00138 to 0.00137, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0024 - val_loss: 0.0014
Epoch 48/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0024Epoch 00047: val_loss improved from 0.00137 to 0.00137, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0024 - val_loss: 0.0014
Epoch 49/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0023Epoch 00048: val_loss improved from 0.00137 to 0.00134, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0024 - val_loss: 0.0013
Epoch 50/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0023Epoch 00049: val_loss improved from 0.00134 to 0.00130, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0023 - val_loss: 0.0013
Epoch 51/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0023Epoch 00050: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0023 - val_loss: 0.0013
Epoch 52/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0023Epoch 00051: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0023 - val_loss: 0.0014
Epoch 53/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0023Epoch 00052: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0023 - val_loss: 0.0013
Epoch 54/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0022Epoch 00053: val_loss improved from 0.00130 to 0.00126, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0022 - val_loss: 0.0013
Epoch 55/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0022Epoch 00054: val_loss improved from 0.00126 to 0.00126, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0022 - val_loss: 0.0013
Epoch 56/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0021Epoch 00055: val_loss improved from 0.00126 to 0.00124, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0021 - val_loss: 0.0012
Epoch 57/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0021Epoch 00056: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0021 - val_loss: 0.0012
Epoch 58/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0020Epoch 00057: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0020 - val_loss: 0.0013
Epoch 59/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0020Epoch 00058: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0020 - val_loss: 0.0013
Epoch 60/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0021Epoch 00059: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0021 - val_loss: 0.0013
Epoch 61/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0020Epoch 00060: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0020 - val_loss: 0.0014
Epoch 62/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0020Epoch 00061: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0020 - val_loss: 0.0012
Epoch 63/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0020Epoch 00062: val_loss improved from 0.00124 to 0.00118, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0020 - val_loss: 0.0012
Epoch 64/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0019Epoch 00063: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0019 - val_loss: 0.0012
Epoch 65/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0019Epoch 00064: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0019 - val_loss: 0.0013
Epoch 66/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0019Epoch 00065: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0019 - val_loss: 0.0012
Epoch 67/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0019Epoch 00066: val_loss improved from 0.00118 to 0.00109, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0019 - val_loss: 0.0011
Epoch 68/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0019Epoch 00067: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0019 - val_loss: 0.0013
Epoch 69/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0018Epoch 00068: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0018 - val_loss: 0.0011
Epoch 70/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0018Epoch 00069: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0018 - val_loss: 0.0011
Epoch 71/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0018Epoch 00070: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0018 - val_loss: 0.0012
Epoch 72/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0018Epoch 00071: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0018 - val_loss: 0.0012
Epoch 73/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0017Epoch 00072: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0017 - val_loss: 0.0011
Epoch 74/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0017Epoch 00073: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0017 - val_loss: 0.0011
Epoch 75/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0017Epoch 00074: val_loss improved from 0.00109 to 0.00108, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0017 - val_loss: 0.0011
Epoch 76/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0017Epoch 00075: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0017 - val_loss: 0.0012
Epoch 77/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0017Epoch 00076: val_loss improved from 0.00108 to 0.00105, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0017 - val_loss: 0.0011
Epoch 78/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0017Epoch 00077: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0017 - val_loss: 0.0011
Epoch 79/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0016Epoch 00078: val_loss improved from 0.00105 to 0.00101, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0016 - val_loss: 0.0010
Epoch 80/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0016Epoch 00079: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0016 - val_loss: 0.0010
Epoch 81/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0016Epoch 00080: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0016 - val_loss: 0.0011
Epoch 82/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0016Epoch 00081: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0016 - val_loss: 0.0010
Epoch 83/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0015Epoch 00082: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0015 - val_loss: 0.0011
Epoch 84/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0015Epoch 00083: val_loss improved from 0.00101 to 0.00101, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0016 - val_loss: 0.0010
Epoch 85/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0015Epoch 00084: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0015 - val_loss: 0.0012
Epoch 86/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0015Epoch 00085: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0015 - val_loss: 0.0011
Epoch 87/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0015Epoch 00086: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0015 - val_loss: 0.0011
Epoch 88/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0015Epoch 00087: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0015 - val_loss: 0.0011
Epoch 89/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0015Epoch 00088: val_loss improved from 0.00101 to 0.00098, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0015 - val_loss: 9.7735e-04
Epoch 90/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0014Epoch 00089: val_loss improved from 0.00098 to 0.00097, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0014 - val_loss: 9.6580e-04
Epoch 91/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0014Epoch 00090: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0014 - val_loss: 9.7901e-04
Epoch 92/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0014Epoch 00091: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0014 - val_loss: 0.0010
Epoch 93/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0014Epoch 00092: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0014 - val_loss: 0.0011
Epoch 94/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0014Epoch 00093: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0014 - val_loss: 0.0010
Epoch 95/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00094: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0013 - val_loss: 0.0010
Epoch 96/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00095: val_loss improved from 0.00097 to 0.00090, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0013 - val_loss: 9.0261e-04
Epoch 97/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00096: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0014 - val_loss: 0.0010
Epoch 98/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00097: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0013 - val_loss: 0.0010
Epoch 99/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00098: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0013 - val_loss: 9.2609e-04
Epoch 100/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00099: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0013 - val_loss: 0.0010
Epoch 101/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00100: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0013 - val_loss: 9.8189e-04
Epoch 102/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0013Epoch 00101: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0013 - val_loss: 0.0010
Epoch 103/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00102: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 9.7707e-04
Epoch 104/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00103: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 9.7877e-04
Epoch 105/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00104: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 9.2640e-04
Epoch 106/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00105: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 9.8469e-04
Epoch 107/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00106: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 9.2922e-04
Epoch 108/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00107: val_loss improved from 0.00090 to 0.00088, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 8.8395e-04
Epoch 109/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00108: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 9.2117e-04
Epoch 110/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0012Epoch 00109: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0012 - val_loss: 9.1833e-04
Epoch 111/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00110: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 8.9735e-04
Epoch 112/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00111: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 9.1030e-04
Epoch 113/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00112: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 8.9327e-04
Epoch 114/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00113: val_loss improved from 0.00088 to 0.00087, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 8.7073e-04
Epoch 115/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00114: val_loss improved from 0.00087 to 0.00086, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 8.5548e-04
Epoch 116/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00115: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 9.0159e-04
Epoch 117/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00116: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 8.9647e-04
Epoch 118/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00117: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 9.0986e-04
Epoch 119/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0010Epoch 00118: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0010 - val_loss: 8.9392e-04
Epoch 120/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00119: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 8.7237e-04
Epoch 121/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0011Epoch 00120: val_loss improved from 0.00086 to 0.00084, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0011 - val_loss: 8.4316e-04
Epoch 122/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0010Epoch 00121: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 0.0010 - val_loss: 9.3286e-04
Epoch 123/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.8665e-04Epoch 00122: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.9739e-04 - val_loss: 0.0011
Epoch 124/1000
1664/1712 [============================>.] - ETA: 0s - loss: 0.0010Epoch 00123: val_loss improved from 0.00084 to 0.00083, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 0.0010 - val_loss: 8.3224e-04
Epoch 125/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.8836e-04Epoch 00124: val_loss improved from 0.00083 to 0.00081, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 9.8686e-04 - val_loss: 8.1177e-04
Epoch 126/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.8299e-04Epoch 00125: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.7857e-04 - val_loss: 8.5119e-04
Epoch 127/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.9136e-04Epoch 00126: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.8753e-04 - val_loss: 8.5464e-04
Epoch 128/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.6385e-04Epoch 00127: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.6526e-04 - val_loss: 8.9758e-04
Epoch 129/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.5121e-04Epoch 00128: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.4905e-04 - val_loss: 9.0830e-04
Epoch 130/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.3512e-04Epoch 00129: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.3589e-04 - val_loss: 8.2437e-04
Epoch 131/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.3306e-04Epoch 00130: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.3557e-04 - val_loss: 8.2150e-04
Epoch 132/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.2724e-04Epoch 00131: val_loss improved from 0.00081 to 0.00081, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 9.2960e-04 - val_loss: 8.0636e-04
Epoch 133/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.1317e-04Epoch 00132: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.1324e-04 - val_loss: 8.5627e-04
Epoch 134/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.3326e-04Epoch 00133: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.3054e-04 - val_loss: 8.7163e-04
Epoch 135/1000
1664/1712 [============================>.] - ETA: 0s - loss: 9.0309e-04Epoch 00134: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 9.0059e-04 - val_loss: 8.4806e-04
Epoch 136/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.8318e-04Epoch 00135: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.8354e-04 - val_loss: 8.1481e-04
Epoch 137/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.6718e-04Epoch 00136: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.6710e-04 - val_loss: 8.0958e-04
Epoch 138/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.6887e-04Epoch 00137: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.6483e-04 - val_loss: 8.2489e-04
Epoch 139/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.5477e-04Epoch 00138: val_loss improved from 0.00081 to 0.00080, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 8.5828e-04 - val_loss: 8.0406e-04
Epoch 140/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.7750e-04Epoch 00139: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.7736e-04 - val_loss: 8.8374e-04
Epoch 141/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.4102e-04Epoch 00140: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.3701e-04 - val_loss: 8.7153e-04
Epoch 142/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.5067e-04Epoch 00141: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.5216e-04 - val_loss: 9.0469e-04
Epoch 143/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.2075e-04Epoch 00142: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.2261e-04 - val_loss: 8.7747e-04
Epoch 144/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.1705e-04Epoch 00143: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.1700e-04 - val_loss: 8.1563e-04
Epoch 145/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.0509e-04Epoch 00144: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.0429e-04 - val_loss: 8.3552e-04
Epoch 146/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.9451e-04Epoch 00145: val_loss improved from 0.00080 to 0.00076, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 7.9829e-04 - val_loss: 7.6140e-04
Epoch 147/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.6508e-04Epoch 00146: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.6542e-04 - val_loss: 8.2557e-04
Epoch 148/1000
1664/1712 [============================>.] - ETA: 0s - loss: 8.3589e-04Epoch 00147: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 8.3685e-04 - val_loss: 8.5541e-04
Epoch 149/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.6974e-04Epoch 00148: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.7381e-04 - val_loss: 7.7585e-04
Epoch 150/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.7149e-04Epoch 00149: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.7179e-04 - val_loss: 7.9411e-04
Epoch 151/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.8572e-04Epoch 00150: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.8354e-04 - val_loss: 8.8088e-04
Epoch 152/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.4870e-04Epoch 00151: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.4762e-04 - val_loss: 8.3640e-04
Epoch 153/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.6242e-04Epoch 00152: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.6155e-04 - val_loss: 8.1267e-04
Epoch 154/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.3832e-04Epoch 00153: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.4016e-04 - val_loss: 8.1383e-04
Epoch 155/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.6677e-04Epoch 00154: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.6872e-04 - val_loss: 8.5529e-04
Epoch 156/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.3892e-04Epoch 00155: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.3492e-04 - val_loss: 7.9889e-04
Epoch 157/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.3379e-04Epoch 00156: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.3480e-04 - val_loss: 7.8281e-04
Epoch 158/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.6441e-04Epoch 00157: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.6465e-04 - val_loss: 8.8282e-04
Epoch 159/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.3707e-04Epoch 00158: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.3520e-04 - val_loss: 8.1434e-04
Epoch 160/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.1617e-04Epoch 00159: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.1590e-04 - val_loss: 7.6970e-04
Epoch 161/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.2166e-04Epoch 00160: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.2358e-04 - val_loss: 7.8296e-04
Epoch 162/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.0994e-04Epoch 00161: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.0961e-04 - val_loss: 8.0845e-04
Epoch 163/1000
1664/1712 [============================>.] - ETA: 0s - loss: 7.0086e-04Epoch 00162: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 7.0799e-04 - val_loss: 7.9203e-04
Epoch 164/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.8663e-04Epoch 00163: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.8601e-04 - val_loss: 8.3451e-04
Epoch 165/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.8596e-04Epoch 00164: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.8708e-04 - val_loss: 7.7569e-04
Epoch 166/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.7686e-04Epoch 00165: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.7705e-04 - val_loss: 7.6205e-04
Epoch 167/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.7750e-04Epoch 00166: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.7677e-04 - val_loss: 8.0213e-04
Epoch 168/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.7601e-04Epoch 00167: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.7569e-04 - val_loss: 7.9878e-04
Epoch 169/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.7992e-04Epoch 00168: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.7899e-04 - val_loss: 7.9351e-04
Epoch 170/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.7723e-04Epoch 00169: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.7669e-04 - val_loss: 8.0375e-04
Epoch 171/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.4698e-04Epoch 00170: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.4746e-04 - val_loss: 7.9470e-04
Epoch 172/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.5193e-04Epoch 00171: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.5124e-04 - val_loss: 8.0527e-04
Epoch 173/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.4748e-04Epoch 00172: val_loss improved from 0.00076 to 0.00074, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 6.4779e-04 - val_loss: 7.4362e-04
Epoch 174/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.5605e-04Epoch 00173: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.5691e-04 - val_loss: 7.7102e-04
Epoch 175/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.4550e-04Epoch 00174: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.4342e-04 - val_loss: 7.4366e-04
Epoch 176/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.3984e-04Epoch 00175: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.4124e-04 - val_loss: 7.6149e-04
Epoch 177/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.2459e-04Epoch 00176: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.2385e-04 - val_loss: 8.1050e-04
Epoch 178/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.2676e-04Epoch 00177: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.2222e-04 - val_loss: 7.9223e-04
Epoch 179/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.1636e-04Epoch 00178: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.2011e-04 - val_loss: 8.2725e-04
Epoch 180/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.1145e-04Epoch 00179: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.1185e-04 - val_loss: 7.5728e-04
Epoch 181/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.2109e-04Epoch 00180: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.2158e-04 - val_loss: 8.2278e-04
Epoch 182/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.0480e-04Epoch 00181: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.0774e-04 - val_loss: 7.6695e-04
Epoch 183/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.9866e-04Epoch 00182: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.9796e-04 - val_loss: 7.6470e-04
Epoch 184/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.0267e-04Epoch 00183: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.0005e-04 - val_loss: 7.8333e-04
Epoch 185/1000
1664/1712 [============================>.] - ETA: 0s - loss: 6.0339e-04Epoch 00184: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 6.0528e-04 - val_loss: 8.4298e-04
Epoch 186/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.9521e-04Epoch 00185: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.9523e-04 - val_loss: 7.4710e-04
Epoch 187/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.7596e-04Epoch 00186: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.7664e-04 - val_loss: 7.6596e-04
Epoch 188/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.7593e-04Epoch 00187: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.7709e-04 - val_loss: 7.6194e-04
Epoch 189/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.8380e-04Epoch 00188: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.8472e-04 - val_loss: 8.3751e-04
Epoch 190/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.7961e-04Epoch 00189: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.8127e-04 - val_loss: 7.6124e-04
Epoch 191/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.7831e-04Epoch 00190: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.7624e-04 - val_loss: 7.4866e-04
Epoch 192/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.8059e-04Epoch 00191: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.7985e-04 - val_loss: 7.8653e-04
Epoch 193/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.5731e-04Epoch 00192: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.6089e-04 - val_loss: 7.9028e-04
Epoch 194/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.7184e-04Epoch 00193: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.7154e-04 - val_loss: 8.3041e-04
Epoch 195/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.6032e-04Epoch 00194: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.5754e-04 - val_loss: 7.4810e-04
Epoch 196/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.5057e-04Epoch 00195: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.4791e-04 - val_loss: 8.1656e-04
Epoch 197/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.4091e-04Epoch 00196: val_loss improved from 0.00074 to 0.00073, saving model to saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5
1712/1712 [==============================] - 4s - loss: 5.3968e-04 - val_loss: 7.2988e-04
Epoch 198/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.3517e-04Epoch 00197: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.3288e-04 - val_loss: 7.5641e-04
Epoch 199/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.3486e-04Epoch 00198: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.3525e-04 - val_loss: 8.2274e-04
Epoch 200/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.3620e-04Epoch 00199: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.3735e-04 - val_loss: 8.2869e-04
Epoch 201/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.4488e-04Epoch 00200: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.4602e-04 - val_loss: 7.8114e-04
Epoch 202/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.2975e-04Epoch 00201: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.2909e-04 - val_loss: 7.3330e-04
Epoch 203/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.2984e-04Epoch 00202: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.2903e-04 - val_loss: 8.0530e-04
Epoch 204/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.3000e-04Epoch 00203: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.3110e-04 - val_loss: 7.7504e-04
Epoch 205/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.2230e-04Epoch 00204: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.1997e-04 - val_loss: 7.5687e-04
Epoch 206/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.2772e-04Epoch 00205: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.2950e-04 - val_loss: 7.4863e-04
Epoch 207/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.1826e-04Epoch 00206: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.2024e-04 - val_loss: 7.4596e-04
Epoch 208/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.2309e-04Epoch 00207: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.2099e-04 - val_loss: 7.5519e-04
Epoch 209/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0143e-04Epoch 00208: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0342e-04 - val_loss: 7.4521e-04
Epoch 210/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0780e-04Epoch 00209: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0751e-04 - val_loss: 7.4769e-04
Epoch 211/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0873e-04Epoch 00210: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0823e-04 - val_loss: 7.3264e-04
Epoch 212/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.1270e-04Epoch 00211: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.1383e-04 - val_loss: 7.5708e-04
Epoch 213/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0110e-04Epoch 00212: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0121e-04 - val_loss: 7.8078e-04
Epoch 214/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.1232e-04Epoch 00213: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.1352e-04 - val_loss: 7.6321e-04
Epoch 215/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0385e-04Epoch 00214: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0205e-04 - val_loss: 7.3336e-04
Epoch 216/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0290e-04Epoch 00215: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0384e-04 - val_loss: 7.4026e-04
Epoch 217/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0092e-04Epoch 00216: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0009e-04 - val_loss: 7.4244e-04
Epoch 218/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.9961e-04Epoch 00217: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.9974e-04 - val_loss: 7.7327e-04
Epoch 219/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0822e-04Epoch 00218: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0817e-04 - val_loss: 7.8244e-04
Epoch 220/1000
1664/1712 [============================>.] - ETA: 0s - loss: 5.0375e-04Epoch 00219: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 5.0151e-04 - val_loss: 7.9479e-04
Epoch 221/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.9529e-04Epoch 00220: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.9373e-04 - val_loss: 7.6674e-04
Epoch 222/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.8653e-04Epoch 00221: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.8604e-04 - val_loss: 7.7566e-04
Epoch 223/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.8353e-04Epoch 00222: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.8412e-04 - val_loss: 7.6314e-04
Epoch 224/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.8231e-04Epoch 00223: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.8265e-04 - val_loss: 7.4425e-04
Epoch 225/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.7676e-04Epoch 00224: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.7852e-04 - val_loss: 7.7305e-04
Epoch 226/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.7407e-04Epoch 00225: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.7527e-04 - val_loss: 7.8584e-04
Epoch 227/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.7304e-04Epoch 00226: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.7289e-04 - val_loss: 7.7512e-04
Epoch 228/1000
1664/1712 [============================>.] - ETA: 0s - loss: 4.6974e-04Epoch 00227: val_loss did not improve
1712/1712 [==============================] - 4s - loss: 4.6864e-04 - val_loss: 7.6661e-04

Step 7: Visualize the Loss and Test Predictions

(IMPLEMENTATION) Answer a few questions and visualize the loss

Question 1: Outline the steps you took to get to your final neural network architecture and your reasoning at each step.

Answer: Creating the neural network architecture was probably the most challenging task of this project. On the one hand there is lots of material that can be used and transferred to our problem, on the other hand there are not yet rules that define how a good architecture for a given problem can be accomplished. Since most of our current knowledge relies on empirical approaches, it is very difficult to predict wether a specific archtitecture will work for a problem or if it has to be entirely redesigned. Before I finally got my CNN architecture, I experimented with many different ideas. Some of them really didn't work and have been dismissed wheras other showed some benefits and have been further improved. Basically I followed and Steps:

  1. Review of related course material
  2. Investigation of common CNN Architectures which are applicable to this problem
  3. Defining the keras sequential model as a basis for the next steps
  4. Adding a Convolutional Layer with an input that corresponds to the training data
  5. Adding hidden layers
  6. Adding a Dense Layer that corresponds to the output of the training data
  7. Compiling and training the model on a small batch size to valdating that model can fit data
  8. Plotting of train_loss and validation_loss

After this I created several prototypes off CNN architectures by adding hidden layers to the models sequence. Within the model that achieved a combination of good results and performance, I tried to apply some of the building blocks of the VGGNet (see image).

VGG

In my final model I use Kernels of the size (3x3) and filters with sizes 16,32,64. The first two blocks have 2 Convolutional Layers that are followed by a MaxPooling and a Dropout Layer. The third blocks deviates from this pattern because both of the Convolutional Layers are followed by a MaxPooling Layer. I noticed that this measure drastically reduced the number of total paramaters and proofed to be a good trade off. Finally the model is completed by block consisting out of a Flatten, a Dense and a Dropout Layer.

Question 2: Defend your choice of optimizer. Which optimizers did you test, and how did you determine which worked best?

Answer: The optimizer I finally have choosen is AdaMax which is a variant of Adam based on the infinity norm. https://arxiv.org/pdf/1412.6980.pdf . The method combines the advantages of two recently popular optimization methods: the ability of AdaGrad to deal with sparse gradients, and the ability of RMSProp to deal with non-stationary objectives.

VGG

I tested the following optimizer: SGD, RMSprop, Adagrad, Adadelta, Adam, Adamax, Nadam. In the following you can find the results for AdaMax. For choosing the best option I compared their rates of optimizing train and validation loss.

AdaMax.png

Use the code cell below to plot the training and validation loss of your neural network. You may find this resource useful.

In [184]:
## TODO: Visualize the training and validation loss of your neural network

plt.style.use('ggplot')
plt.figure(figsize=(20,10))
plt.plot(hist.history['loss'],'tab:green')
plt.plot(hist.history['val_loss'],'tab:red')
plt.title('model loss'+'    ['+ detail + ']')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['loss', 'val_loss'], loc='upper right')
plt.savefig('saved_models/{}_{}.png'.format(detail, time.time()))
plt.show()

Question 3: Do you notice any evidence of overfitting or underfitting in the above plot? If so, what steps have you taken to improve your model? Note that slight overfitting or underfitting will not hurt your chances of a successful submission, as long as you have attempted some solutions towards improving your model (such as regularization, dropout, increased/decreased number of layers, etc).

Answer: In the above plot one can identify slight signs of overfitting as the validation loss at some point (~epoch 150) does not decrease any further although the optimizer continues to minimize the loss function. This means the model is optimized to much to represent the training data and fails to generalize to unseen data. In order to improve the model, I removed an additional dense layer from a previous version build and also increased the dropout rate between the last two layers. Another option to further improve the model and in addition to reduce overfitting is to apply data augmentation. (Unfortunately, I exceeded my aws credit for gpu computing so I could not continue to investigate.)

Visualize a Subset of the Test Predictions

Execute the code cell below to visualize your model's predicted keypoints on a subset of the testing images.

In [164]:
y_test = model.predict(X_test)
fig = plt.figure(figsize=(20,20))
fig.subplots_adjust(left=0, right=1, bottom=0, top=1, hspace=0.05, wspace=0.05)
for i in range(9):
    ax = fig.add_subplot(3, 3, i + 1, xticks=[], yticks=[])
    plot_data(X_test[i], y_test[i], ax)

Step 8: Complete the pipeline

With the work you did in Sections 1 and 2 of this notebook, along with your freshly trained facial keypoint detector, you can now complete the full pipeline. That is given a color image containing a person or persons you can now

  • Detect the faces in this image automatically using OpenCV
  • Predict the facial keypoints in each face detected in the image
  • Paint predicted keypoints on each face detected

In this Subsection you will do just this!

(IMPLEMENTATION) Facial Keypoints Detector

Use the OpenCV face detection functionality you built in previous Sections to expand the functionality of your keypoints detector to color images with arbitrary size. Your function should perform the following steps

  1. Accept a color image.
  2. Convert the image to grayscale.
  3. Detect and crop the face contained in the image.
  4. Locate the facial keypoints in the cropped image.
  5. Overlay the facial keypoints in the original (color, uncropped) image.

Note: step 4 can be the trickiest because remember your convolutional network is only trained to detect facial keypoints in $96 \times 96$ grayscale images where each pixel was normalized to lie in the interval $[0,1]$, and remember that each facial keypoint was normalized during training to the interval $[-1,1]$. This means - practically speaking - to paint detected keypoints onto a test face you need to perform this same pre-processing to your candidate face - that is after detecting it you should resize it to $96 \times 96$ and normalize its values before feeding it into your facial keypoint detector. To be shown correctly on the original image the output keypoints from your detector then need to be shifted and re-normalized from the interval $[-1,1]$ to the width and height of your detected face.

When complete you should be able to produce example images like the one below

In [5]:
# Load in color image for face detection
image = cv2.imread('images/obamas4.jpg')


# Convert the image to RGB colorspace
image_copy = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)


# plot our image
fig = plt.figure(figsize = (9,9))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])
ax1.set_title('image copy')
ax1.imshow(image_copy)

from keras.models import load_model

model = load_model('saved_models/updated_best_model_mean_squared_error_adamax_1515354036.5151918.h5')
In [6]:
### TODO: Use the face detection code we saw in Section 1 with your trained conv-net 
## TODO : Paint the predicted keypoints on the test image

def predict_face_keypoints(image, face_cascade = cv2.CascadeClassifier('detector_architectures/haarcascade_frontalface_default.xml')):
    
    original = image
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
       
    for (x,y,w,h) in faces:
        
        cv2.rectangle(original,(x,y),(x+w,y+h),(255,0,0),1)    
        
        face_crop = image[y:y+h, x:x+w]
        shape_crop = face_crop.shape
        gray_crop = cv2.cvtColor(face_crop, cv2.COLOR_BGR2GRAY)
        shrinked_normalized_gray_crop = cv2.resize(gray_crop, (96, 96)) / 255.
        shrinked_landmarks = np.squeeze(model.predict(np.expand_dims(np.expand_dims(shrinked_normalized_gray_crop, axis=-1), axis=0)))
        projected_landmarks_x = (shrinked_landmarks[0::2] * 48 + 48)*shape_crop[0]/96
        projected_landmarks_y = (shrinked_landmarks[1::2] * 48 + 48)*shape_crop[1]/96
        for x_pred,y_pred in zip(projected_landmarks_x,projected_landmarks_y):
            cv2.circle(original, (int(x+x_pred),int(y+y_pred)), 1, (0,255,0), thickness=-1, lineType=cv2.LINE_AA)
        
    return original

plt.figure(figsize = (17,17))
plt.imshow(predict_face_keypoints(image_copy))
Out[6]:
<matplotlib.image.AxesImage at 0x1dbead24ac8>

(Optional) Further Directions - add a filter using facial keypoints to your laptop camera

Now you can add facial keypoint detection to your laptop camera - as illustrated in the gif below.

The next Python cell contains the basic laptop video camera function used in the previous optional video exercises. Combine it with the functionality you developed for keypoint detection and marking in the previous exercise and you should be good to go!

In [7]:
import cv2
import time 
from keras.models import load_model



def laptop_camera_go():
    # Create instance of video capturer
    cv2.namedWindow("face detection activated")
    vc = cv2.VideoCapture(0)

    # Try to get the first frame
    if vc.isOpened(): 
        rval, frame = vc.read()
    else:
        rval = False
    
    # keep video stream open
    while rval:
        
        
        # plot image from camera with detections marked
        cv2.imshow("face detection activated", predict_face_keypoints(frame))
        
        # exit functionality - press any key to exit laptop video
        key = cv2.waitKey(20)
        if key > 0: # exit by pressing any key
            # destroy windows
            cv2.destroyAllWindows()
            
            # hack from stack overflow for making sure window closes on osx --> https://stackoverflow.com/questions/6116564/destroywindow-does-not-close-window-on-mac-using-python-and-opencv
            for i in range (1,5):
                cv2.waitKey(1)
            return
        
        # read next frame
        time.sleep(0.05)             # control framerate for computation - default 20 frames per sec
        rval, frame = vc.read()  
In [8]:
# Run your keypoint face painter
laptop_camera_go()

(Optional) Further Directions - add a filter using facial keypoints

Using your freshly minted facial keypoint detector pipeline you can now do things like add fun filters to a person's face automatically. In this optional exercise you can play around with adding sunglasses automatically to each individual's face in an image as shown in a demonstration image below.

To produce this effect an image of a pair of sunglasses shown in the Python cell below.

In [6]:
# Load in sunglasses image - note the usage of the special option
# cv2.IMREAD_UNCHANGED, this option is used because the sunglasses 
# image has a 4th channel that allows us to control how transparent each pixel in the image is
sunglasses = cv2.imread("images/sunglasses_4.png", cv2.IMREAD_UNCHANGED)

# Plot the image
fig = plt.figure(figsize = (6,6))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])
ax1.imshow(sunglasses)
ax1.axis('off');

This image is placed over each individual's face using the detected eye points to determine the location of the sunglasses, and eyebrow points to determine the size that the sunglasses should be for each person (one could also use the nose point to determine this).

Notice that this image actually has 4 channels, not just 3.

In [7]:
# Print out the shape of the sunglasses image
print ('The sunglasses image has shape: ' + str(np.shape(sunglasses)))
The sunglasses image has shape: (1123, 3064, 4)

It has the usual red, blue, and green channels any color image has, with the 4th channel representing the transparency level of each pixel in the image. Here's how the transparency channel works: the lower the value, the more transparent the pixel will become. The lower bound (completely transparent) is zero here, so any pixels set to 0 will not be seen.

This is how we can place this image of sunglasses on someone's face and still see the area around of their face where the sunglasses lie - because these pixels in the sunglasses image have been made completely transparent.

Lets check out the alpha channel of our sunglasses image in the next Python cell. Note because many of the pixels near the boundary are transparent we'll need to explicitly print out non-zero values if we want to see them.

In [8]:
# Print out the sunglasses transparency (alpha) channel
alpha_channel = sunglasses[:,:,3]
print ('the alpha channel here looks like')
print (alpha_channel)

# Just to double check that there are indeed non-zero values
# Let's find and print out every value greater than zero
values = np.where(alpha_channel != 0)
print ('\n the non-zero values of the alpha channel look like')
print (values)
the alpha channel here looks like
[[0 0 0 ..., 0 0 0]
 [0 0 0 ..., 0 0 0]
 [0 0 0 ..., 0 0 0]
 ..., 
 [0 0 0 ..., 0 0 0]
 [0 0 0 ..., 0 0 0]
 [0 0 0 ..., 0 0 0]]

 the non-zero values of the alpha channel look like
(array([  17,   17,   17, ..., 1109, 1109, 1109], dtype=int64), array([ 687,  688,  689, ..., 2376, 2377, 2378], dtype=int64))

This means that when we place this sunglasses image on top of another image, we can use the transparency channel as a filter to tell us which pixels to overlay on a new image (only the non-transparent ones with values greater than zero).

One last thing: it's helpful to understand which keypoint belongs to the eyes, mouth, etc. So, in the image below, we also display the index of each facial keypoint directly on the image so that you can tell which keypoints are for the eyes, eyebrows, etc.

With this information, you're well on your way to completing this filtering task! See if you can place the sunglasses automatically on the individuals in the image loaded in / shown in the next Python cell.

In [9]:
# Load in color image for face detection
image = cv2.imread('images/obamas4.jpg')

# Convert the image to RGB colorspace
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)


# Plot the image
fig = plt.figure(figsize = (8,8))
ax1 = fig.add_subplot(111)
ax1.set_xticks([])
ax1.set_yticks([])
ax1.set_title('Original Image')
ax1.imshow(image)
Out[9]:
<matplotlib.image.AxesImage at 0x28d60c827f0>
In [ ]:
## (Optional) TODO: Use the face detection code we saw in Section 1 with your trained conv-net to put
## sunglasses on the individuals in our test image

(Optional) Further Directions - add a filter using facial keypoints to your laptop camera

Now you can add the sunglasses filter to your laptop camera - as illustrated in the gif below.

The next Python cell contains the basic laptop video camera function used in the previous optional video exercises. Combine it with the functionality you developed for adding sunglasses to someone's face in the previous optional exercise and you should be good to go!

In [ ]:
import cv2
import time 
from keras.models import load_model
import numpy as np

def laptop_camera_go():
    # Create instance of video capturer
    cv2.namedWindow("face detection activated")
    vc = cv2.VideoCapture(0)

    # try to get the first frame
    if vc.isOpened(): 
        rval, frame = vc.read()
    else:
        rval = False
    
    # Keep video stream open
    while rval:
        # Plot image from camera with detections marked
        cv2.imshow("face detection activated", frame)
        
        # Exit functionality - press any key to exit laptop video
        key = cv2.waitKey(20)
        if key > 0: # exit by pressing any key
            # Destroy windows 
            cv2.destroyAllWindows()
            
            for i in range (1,5):
                cv2.waitKey(1)
            return
        
        # Read next frame
        time.sleep(0.05)             # control framerate for computation - default 20 frames per sec
        rval, frame = vc.read()    
        
In [ ]:
# Load facial landmark detector model
model = load_model('my_model.h5')

# Run sunglasses painter
laptop_camera_go()